Lecture【DOC精选】.doc

Lecture 18 Integer Programming 12.8 If the LP relaxation is infeasible, then the ILP is infeasible. 12.10 If the optimal solution to the LP is integer, then this is the optimal solution to the ILP also. Problem P: max cx s.t. Ax = b 0 x u xj integer for all j Problem LP: max cx s.t. Ax = b 0 x u Let x1 be feasible for P. Then cx1 is a lower bound for problem P. Let x2 be an optimum for LP. Then cx2 is an upper bound for P. gap = (120-100)/120 = 16.7% When constructing an LP model, we usually try to use the smallest number of constraints and variables. ----------------------------------------------- When constructing an ILP model, we usually try to make the optimal objective value of the LP relaxation as small as possible. This may involve using more constraints than is absolutely necessary. Consider a problem with constraints of the form: select x or y, select x or z, select y or z max x+y+z s.t. x + y 1 x + z 1 y + z 1 x, y, z binary LP relaxation yields (1/2, 1/2, 1/2) with objective value of 3/2. -------------------------------------------- max x+y+z s.t. x + y 1 x + z 1 y + z 1 x + y + z 1 x, y, z binary LP relaxation yields (1,0,0) with objective value of 1 (a better relaxation). We used a redundant constraint to obtain an improved relaxation. The gap between the LP and ILP is smaller. Im working on a telecom design problem where customers for wireless phone service are assigned to towers and channels are assigned to customer-tower combinations. Assumptions 1. Customers can be assigned to either tower 2. tower #1 capacity = 800 customers, cost = $300,000 3. tower #2 capacity = 750 customers, cost = $288,888 4. revenue per customer each year = $750 ----------------------------------------------- Determine which tower(s) to build and which customers to assign to the towers. Run # 1 printfModel1\n\n; var C1T1 = 0; var C1T2 = 0; var C2T1 = 0; var C2T2 = 0